The overall goal of this research is to vertically integrate detailed knowledge about synapses and other neural elements into an overall understanding of what neural circuits do and how they are designed to do it. An important constraint on the design of a sensory circuit is noise generated by the physical stimulus and by elements of the circuit. This is epitomized by the rod synapse in the retina, whose design removes transduction noise generated within the rod. Another good example from retina is gap junctions that open and close according to the competing demands of limiting photon noise and improving spatial resolution. Also in the retina, multiple synapses converge on a single ganglion cell to average out synaptic noise. ! The present focus of this research is the synapse that the retinal bipolar cell makes on ganglion and amacrine cells. This synapse has an intriguing dyadic structure that brings two neurons postsynaptic to the same vesicle fusion site. Yet the purpose of this structure is not yet clear. At the mechanistic level, because vesicle fusions are random, they induce noise in the postsynaptic neuron. If the dyadic structure includes two postsynaptic ganglion cells, the synapse should induce noise correlations between them. If the postsynaptic neurons are a ganglion cell and an inhibitory amacrine cell that feeds forward to the ganglion cell, the synapse should induce correlations between excitatory and inhibitory currents within the ganglion cell. At the systems level, noise correlations between and within ganglion cells optimize the detection and discrimination of visual stimuli. To vertically integration these different levels requires experimental evidence that the two postsynaptic processes sense the same vesicle fusion event, and that the dyadic synapse induces noise correlations. ! Therefore the proposed research will investigate the mechanisms and structure of the dyadic synapse. To vertically integrate different levels of knowledge, it will include this mechanistic and structural information in biophysically accurate models of the retina circuit. The models will help test ideas about how dyadic synapses and other circuit elements generate noise correlations that optimize the encoding of visual information. ! Prosthetic visual devices need far more space and energy than the retina they are designed to supplement or replace. As we attempt to miniaturize and optimize prosthetic neural devices, the retina has much to teach us. As prosthetic devices approach the compactness of the neural retina, they will encounter the same problem that the retina does for the same reason: using a small number of stochastic elements to encode information invariably produces noise. No prosthetic device has equaled the retina's sensitivity or resolution; therefore, by understanding how the retina is designed to optimize vision in the face of noise, better prosthetic devices can be designed.